Submitted text
This build sends text to its same-origin Kannuki parser. It does not implement accounts or workspace history. The hosted data model and retention policy remain design decisions.
Sources and system behavior
Reference information for the current Kannuki reader build.
Device-local state
The browser stores display preferences. It does not store analyzed passages or implicit reading history.
Issue reports
A report is created only after confirmation and contains the submitted passage, selected span, and parser evidence. Reports are not training or Gold data.
Parser evidence
Analysis schema v2 preserves lexical units beneath natural display groups and adds non-destructive grammar, expression, named-entity, and compound annotations.
SOURCE LEDGER
Where the visible information comes from
JMdict
Japanese lexical entries, readings, parts of speech, and English glosses.
Project page ↗CC BY-SA 4.0Kannuki rules
Project-authored factual grammar, expression, compound, and fictional named-entity fixtures.
Versioned runtime assetsParser facts, not copied lessonsComparison packet
Project-authored candidate sentences retained for internal presentation and regression tooling.
24 non-Gold casesNot Reader contentDiscovery inventories
JLPT Sensei and private flashcards may identify grammar facts only. Their definitions, explanations, translations, and examples are not Kannuki content.
Pattern discovery onlyPrivate Anki read-onlyAnalysis schema2
Segmentation policy1.1.0
Reader stageM5.2
AccessAnonymous local
Current limitation
Parser output can be wrong. Confidence is diagnostic and is not fully calibrated against the planned Gold corpus.